Skip to content

Latest commit

 

History

History
94 lines (71 loc) · 4.38 KB

README.md

File metadata and controls

94 lines (71 loc) · 4.38 KB
Raw

Datasets and their Usage

The datasets used in the 2DMatGMM project are currently hosted privately. They will be made available on Zenodo soon.

Download links are available here:

GMM Detector Dataset

The GMM Detector Dataset encompasses images of Graphene and WSe2 material flakes, annotated using the labelme tool.

Dataset Training Images Testing Images Annotated Flakes
Graphene 425 1362 1 to 4 layers
WSe2 92 420 1 to 3 layers
Material Class Training Instances (> 300 px) Testing Instances (> 300 px)
Graphene 1 240 938
Graphene 2 239 914
Graphene 3 191 612
Graphene 4 96 494
WSe2 1 76 278
WSe2 2 72 224
WSe2 3 58 171

Structure

The GMM Detector Dataset should follow this directory structure:

GMMDetectorDatasets
├───Graphene
│   ├───annotations
│   ├───test_images
│   ├───test_semantic_masks
│   ├───train_images
│   └───train_semantic_masks
└───WSe2
    ├───annotations
    ├───test_images
    ├───test_semantic_masks
    ├───train_images
    └───train_semantic_masks

The annotations folder contains annotation files in the COCO format, the suffix _200 indicates the minimum number of pixels used in that file, the annotation files with the _full suffix contain all possible flake instances but is not used during evaluation, the evaluation uses the _300 file.

Furthermore the instances described in the COCO annotation file are transcribed as a semantic mask in the semantic mask folder.

Please note, the provided semantic masks only include instances with an area larger than 200px. Given the images are captured with a 20x Objective, this equates to approximately 30μm² in size.

False Positive Detector Dataset

The False Positive Detector Dataset consists of masks from a variety of objects detected by the GMM detector. These instances were manually classified as either a true positive or false positive. These annotations are saved in the annotations.json file.

The dataset includes 1929 training images and 579 testing images.

Split True Positives (1) False Positives (0) Total
Train 1151 778 1929
Test 328 251 579

Structure

The False Positive Detector Dataset should adhere to the following structure:

FalsePositiveDetectorDataset
├───train
│   ├───annotations.json
│   ├───masks
│   └───images
└───test
    ├───annotations.json
    ├───masks
    └───images

The annotations.json file should be structured in such a way that each key is a name of a mask in the masks directory without the extension. The value of each key should be either a 0 or a 1 depending on whether the mask is a false positive or not, respectively. An example of the annotations.json file is shown below:

{
    "eb62e102-9a1d-4bf4-811e-e5d15c8268db": 1,
    "562851ba-1661-470f-98fe-7f937732e77d": 0,
    ...
    "a8920cb6-f7b7-44e1-a921-14de4cef3049": 0,
}

Please note, the images directory is only used for visualization purposes and does not factor into the training process.